The short version: OpenAI’s ChatGPT workspace agents, the tool that lets your team hand off multi-step jobs like drafting a weekly report or triaging support tickets, have been free to run since they launched. That free ride ends July 6, 2026. After that, each agent run draws from a purchased credit pool, with typical tasks costing somewhere between 5 and 25 credits depending on how much text they read and write. Your everyday ChatGPT usage, search, files, canvas, chat, is not affected. Only the autonomous, multi-step agent runs are.
What are ChatGPT workspace agents?
Workspace agents are OpenAI’s answer to the question every small business owner eventually asks about AI: can it just handle the whole task, not just answer one question? Unlike a regular ChatGPT conversation, a workspace agent can be assigned an ongoing job, given access to your company’s tools, and left to work in the background, even while you are offline. OpenAI describes them as capable of preparing reports, writing code, and responding to messages using your existing tools and permissions, without step-by-step supervision.
In practice, the examples OpenAI has shown are refreshingly unglamorous. A software-request agent that triages IT tickets and routes approvals. A feedback-routing agent that pulls signals out of Slack and support channels and turns them into a weekly product summary. A metrics agent that pulls Friday’s numbers, builds the charts, and drafts the narrative before your team meeting starts. None of that is flashy. All of it is the kind of recurring, half-day task that eats a small team’s week.
Workspace agents became generally available on Business, Enterprise, and Edu plans on May 22, 2026, according to OpenAI’s help center, after an earlier research preview. You can trigger them manually, put them on a schedule, or deploy them into Slack so they pick up requests as they arrive.
Why is ChatGPT workspace agents pricing changing on July 6?
Nothing about the product is changing on July 6. What changes is who pays for the compute it uses. Agent runs were free during the preview period, and OpenAI had already pushed that free window back once before extending it again to July 6, 2026, based on reporting at the time the extension was announced. After that date, running an agent inside ChatGPT draws credits from a pool your workspace purchases separately from your per-seat subscription.
That two-step extension is itself a useful signal. It suggests OpenAI was watching adoption closely and wanted more real usage data, and more time for businesses to build habits around agents, before asking anyone to pay for them. If you have been on the fence about trying workspace agents, the fact that OpenAI kept the free period open this long is a decent hint that they want you actually using the thing, not just glancing at it. It also fits a pattern this summer: Microsoft folded Copilot into its core 365 pricing on July 1, and OpenAI’s own memory upgrade reached free ChatGPT users just weeks ago. The big AI vendors are all racing to get small businesses using more of the product before they start charging for it.
How much will a workspace agent actually cost my business?
Per OpenAI’s published rate card, a typical end-to-end agent run built on GPT-5.5 consumes somewhere between 5 and 25 credits, depending on task complexity and how much text the agent has to read and generate. OpenAI’s own worked example, an agent processing roughly 20,000 input tokens, 80,000 cached input tokens, and 5,000 output tokens, comes out to about 7.25 credits. Credits are purchased as a pooled add-on on top of the standard ChatGPT Business seat price, shared across your whole workspace, and valid for 12 months once bought.
Two details matter more than the credit math itself. First, core ChatGPT features, chat, search, file uploads, canvas, are not touched by this change; you only spend credits when an agent runs autonomously. Second, agent runs triggered from outside ChatGPT itself, such as an agent responding directly in a Slack channel, remain in free preview for now. So a support-ticket agent living in Slack keeps running free even after July 6, while the same agent triggered manually inside the ChatGPT app starts drawing credits.
What should a small business actually do before July 6?
Use the days you have left. Run your candidate agent, whether it is the metrics report, the ticket triage, or the feedback digest, several times before the free window closes and watch how it behaves. You cannot yet know your exact dollar cost per credit from public pricing alone, but you can learn the two things that actually determine your bill: how often the task needs to run, and how much text it has to chew through each time. A weekly report that reads a handful of spreadsheets is a very different cost profile than an agent re-reading your entire knowledge base on every trigger.
Once July 6 passes, the practical move is to scope tightly before you scale. Give an agent a narrow, well-defined job with a fixed schedule rather than an open-ended one that fires on every possible trigger, and lean on cached inputs where the tool supports it, since cached tokens cost meaningfully less than fresh ones in OpenAI’s own example. If a task can live entirely inside Slack, it is worth checking whether it still qualifies for the free preview path before you assume it will cost credits. This is the same discipline we found when we looked at how enterprises overspent on AI tools without a proportionate return: the businesses that come out ahead are the ones that define the task before they turn on the automation, not after.
The pricing model here rewards a task you have actually defined well. That is arguably a better forcing function for small teams than a flat monthly agent fee would have been, because it makes you specify the job instead of throwing a vague prompt at an expensive black box and hoping.
Frequently Asked Questions
Do I need to pay for ChatGPT workspace agents starting July 6, 2026?
Only if you run agents from inside the ChatGPT app itself. Regular ChatGPT chat, search, file uploads, and canvas are unaffected. Agent runs triggered outside ChatGPT, such as in a connected Slack channel, remain in free preview for now.
How many credits does a typical workspace agent run cost?
OpenAI’s published rate card puts a typical end-to-end run on GPT-5.5 at roughly 5 to 25 credits, with a worked example of about 7.25 credits for a run involving 20,000 input tokens, 80,000 cached input tokens, and 5,000 output tokens.
What plans include workspace agents?
Workspace agents are available on ChatGPT Business, Enterprise, Edu, and Teachers plans. They became generally available on Business, Enterprise, and Edu on May 22, 2026.
Is it worth trying workspace agents before the free period ends?
Yes. Testing now costs nothing and tells you exactly how many credits your specific use case will actually consume, which is far more useful than estimating from a generic rate card after you are already paying.
For more on how AI pricing and tools are shifting for small businesses this year, see our ongoing AI coverage.
If you run a small team, what is the first task you would actually hand to an agent that runs while you are not watching it? Tell us in the comments, we are collecting real examples for a follow-up piece.
